Objectives: A better understanding of the subnational variations could be paramount to the efficiency and effectiveness of the response to the HIV epidemic. The purpose of this study is to describe the methodology used to produce the first estimates at second subnational level released by UNAIDS. Methods: We selected national population-based surveys with HIV testing and survey clusters geolocation, conducted in 2008 or later. A kernel density estimation approach (prevR) with adaptive bandwidths was used to generate a surface of HIV prevalence. This surface was combined with LandScan global population distribution grid to estimate the spatial distribution of people living with HIV (PLWHIV). Finally, results were adjusted to national UNAIDS's published estimates and merged per second subnational administrative unit. An indicator of the quality of the estimates was computed for each administrative unit. Results: These estimates combine two complementary approaches: the prevR method, focusing on spatial variations of HIV prevalence, as well as national estimates published by UNAIDS, taking into account trends of HIV prevalence over time. Seventeen country reports have been produced. However, quality of the estimates at second subnational level is highly heterogonous between countries, depending on the number of units and the survey sampling size. In some countries, estimates at second subnational level are very uncertain and should be interpreted with caution. Conclusion: These estimates at second subnational level constitute a first step to help countries to better understand their HIV epidemic and to inform programming at lower geographical levels. Further developments are needed to better match local needs.